The AI boom is no longer a “future trend”, it’s here and it’s power-hungry. Hyperscale data centres, Graphic Processing Units (GPU) clusters, colocation and edge data centres… all of them are chewing through serious amounts of electricity in the race to train and run AI models. At the same time, operators are under pressure to improve efficiency, prove sustainability credentials and bill tenants or business units fairly and transparently.
At the heart of all of that is one unglamorous but absolutely critical layer: energy metering and billing.
Why the AI Wave Changes the Energy Metering Game
Data centres have always been heavy energy users, now AI has changed the shape of that demand. High-density GPU racks, liquid cooling systems and more dynamic workloads mean power density per rack has jumped dramatically. It’s not unusual for AI racks to push 30–60 kW and beyond, compared to traditional racks that typically draw 7-15 kW.
AI training runs can create sharp peaks, inference farms can spike around customer traffic patterns and brief power disturbances that might go unnoticed in a traditional IT rack can disrupt a multimillion-dollar training run.
For operators, this translates into three big metering challenges. You need more granularity – whole-of-building metering or basic sub-billing at the floor level is no longer enough. Operators need to know which tenants, pods or zones are responsible for specific loads, how those loads move over time and how they relate to cooling, backup systems and other overheads.
You need better accuracy and reliability, because when a single tenant’s monthly bill is six figures or more, accuracy is non-negotiable. Disputes over kWh, peak demand or power factor penalties can quickly become commercial and legal problems.
You also need metering that can feed intelligent software – the days of “meter as a number on the wall” are over. Modern data centres need energy meters that talk, streaming data to DCIM, EMS, BI tools and customer-facing portals in real time.
What Good Energy Metering Looks Like in AI-Ready Data Centres
So what does a fit-for-purpose energy metering approach actually involve?
Comprehensive Electrical Visibility
A robust design starts with energy meters NATA certified for accuracy & traceability under NMI which are placed at key points in the electrical architecture. This will ensure future proofing for any customer disputes along with government regulatory changes.
Of course any customers that are billed for energy use of electricity in Australia, metering devices must be NMI approved by law.
Key points in the electrical architecture:
Incomer and utility supply, generator and Uninterruptible Power Supply (UPS) outputs
Distribution boards feeding white space
Individual tenant supplies and PDUs
High-density racks where viable
Critical mechanical loads like chilled water systems, cooling units, pumps and cooling towers
This hierarchy lets you reconcile energy from the top down. You can see how much of your total site load is going to IT versus cooling versus other services. You can identify losses and “unallocated” consumption. You can also allocate costs with a clear, auditable chain from the main incomer to each billed entity.
Energy meters with multi-channel capabilities can help reduce panel space and installation cost in retrofit scenarios, a big advantage in older data halls where room is at a premium.
High Quality Measurements & Power Quality Insight
For AI loads, it’s not just about kWh. You also need to understand demand (kW/kVA) to support capacity planning, transformer and UPS sizing and demand charges. Power factor and harmonics matter because AI and GPU loads can introduce harmonic distortion and poor power factor, which impact network capacity and may incur penalties from the utility. You also need visibility into voltage events like sags, swells, transients and unbalance that can stress sensitive equipment.
Energy meters with power quality capabilities, including high-resolution waveform capture, EN/IEC-class accuracy and event logging, give operators the evidence they need to investigate incidents and root causes, justify infrastructure upgrades and demonstrate compliance with connection agreements and internal standards.
In high-value facilities, this level of visibility isn’t a “nice to have”, it’s core risk management.
Communications that Work with your Architecture
Modern data centres are highly instrumented environments. Energy meters must be able to slot cleanly into that ecosystem, typically via standard protocols such as Modbus TCP/IP, Modbus RTU, BACnet/IP or other industry interfaces.
Integration with DCIM, EMS or BMS platforms is essential, along with support for secure, IP-based connectivity for remote monitoring and reporting. The goal is straightforward: one source of truth for energy, demand and power quality, available to both operations teams and commercial stakeholders without manual data wrangling.
Getting Billing Right: from "Guesstimate" to Defensible Invoice
AI-driven data centres frequently support a mix of retail colocation tenants, wholesale customers, internal business units and short-term or “burst” AI compute projects. All of them create costs and all of them need to be billed fairly.
The Basics: kWh plus Overheads
Most data centre billing structures start with kWh consumed at the tenant supply point. On top of that, operators may allocate common services and losses (lighting, security, building services, network rooms), cooling energy (often a significant share of total site usage) and electrical losses in transformers, switchgear and cabling.
Meters at key mechanical and common service loads make it possible to allocate these overheads proportionally, with a transparent basis for the percentages used.
Demand and Capacity Charges for AI Tenants
AI workloads are often spiky and high-density, which raises two important commercial questions: What is the maximum demand the infrastructure must be built for? How do we charge fairly for a tenant who occupies capacity they rarely fully use, but which must be reserved for them?
Here, advanced energy metering data supports more sophisticated billing models, such as demand charges based on kW or kVA peaks over a billing period, reserved capacity fees for guaranteed power envelopes and tiered pricing above certain thresholds (for example, a higher rate for consumption beyond a contracted baseline).
The key is that any such model must be backed by trustworthy energy meter data with clear, understood calculations.
Multi-Site or Multi-Tenant Consolidation
For operators running multiple data centres or multiple pods within a campus, consolidated billing becomes important. A good metering and software combination allows you to roll up usage for a single customer across multiple halls or sites, apply standard tariff structures consistently and provide self-service portals or periodic reports that align to how customers purchase and account for AI compute.
This isn’t just about convenience; it’s part of delivering a premium, enterprise-grade experience that keeps high-value tenants loyal.
Why Software is just as Important as the Energy Meter
High quality meters are the data source. To turn that raw data into something commercially useful, you need energy metering and analytics software that can:
- Collect and store readings from hundreds or thousands of meters
- Apply complex tariff structures including energy, demand, time-of-use and power factor charges
- Allocate shared loads based on agreed formulas
- Generate clear auditable billing reports for tenants and internal stakeholders
- Provide dashboards and trend analysis for capacity planning and efficiency programs.
For AI-heavy data centres in particular, the ability to slice and dice energy data by tenant, zone, rack group, time of day or workload type becomes a major enabler of both cost control and commercial innovation.
It also supports sustainability and ESG reporting: carbon footprint per tenant, per rack or per compute unit becomes possible when you can reliably attribute energy use.
Designing for Retrofit vs Greenfield
Many operators are expanding existing sites to cater for AI workloads, not just building new ones. That raises some practical metering questions.
In retrofit environments, panel space is tight. Multi-circuit or multi-channel meters can dramatically reduce the footprint required to capture multiple circuits. Downtime is unacceptable, so solutions that allow installation with minimal outages or staged cut-overs are essential. Legacy protocols and systems mean energy meters need to interface with existing BMS/DCIM platforms, sometimes across mixed generations of equipment.
In greenfield builds, you have the opportunity to design the metering architecture from day one, aligning it with desired billing models, ESG reporting requirements and future expansion of AI workloads and rack densities. Investing upfront in the right metering and software stack can avoid costly change-outs or work-arounds a few years down the track.
Bringing it all Together
The AI boom is putting unprecedented pressure on data centres – electrically, thermally, commercially and reputationally. Operators are being asked to deliver more power to more dense loads with higher reliability, while proving efficiency and fairness and doing it in a way that scales.
None of that is possible without robust, accurate and integrated energy metering and billing.
A well-designed metering solution does more than just support invoices. It becomes the foundation for:
- Capacity planning and infrastructure investment
- Power quality and risk management
- Cost optimisation for both operator and tenants
- Credible sustainability and ESG reporting
- Better customer experience for demanding AI and cloud clients.
As AI accelerates, the data centre’s energy story is coming into sharper focus. The operators who get their energy metering and billing foundations right today will be the ones best placed to grow with confidence tomorrow.
For Data Centre Operators Navigating the AI Boom
SATEC brings together high-accuracy, power-quality energy metering and flexible software to create a single, reliable source of truth for energy and billing. Our compact, multi-function meters are designed for both new builds and retrofit environments where panel space is tight, giving you detailed visibility from the main incomer right down to tenant feeds, PDUs and critical mechanical loads.
With advanced power quality analysis, demand and harmonics monitoring and NMI-approved accuracy where required, you can trust the numbers that sit behind capacity planning, incident investigation and commercial decisions. Layered over the top, our Expertpower cloud platform turns raw meter data into clear, auditable billing, cost allocation and analytics – whether you’re sub-billing colocation tenants, managing internal AI business units or consolidating usage across multiple sites.
The result is an energy metering and software stack that not only keeps invoices defensible but also supports efficiency, resilience and a better experience for your most demanding data centre customers.
To find out more, talk to our team of experts today.
FAQs - Energy Metering for the Data Centre and AI Boom
Why is energy metering so important for AI-focused data centres?
AI workloads are high-density and often run 24/7, metering is essential to see exactly where power is used, protect capacity margins and maintain uptime.
What’s the difference between basic metering and “power quality” metering?
Basic metering tells you how much energy you’re using, while power quality metering also shows harmonics, voltage dips, unbalance and other issues that can impact sensitive IT and UPS equipment.
Do we need to meter down to individual racks or circuits?
Not everywhere, but for GPU clusters, tenant billing or high-risk areas, circuit-level metering gives the detailed visibility you need for accurate charging and safe capacity planning.
Can advanced metering be added to an existing data centre, or only new builds?
It can absolutely be retrofitted; compact, multi-circuit meters are designed to fit into crowded boards and existing infrastructure with minimal disruption.
